Alleged GeForce RTX Turing "SUPER" Refresh Specs Leaked (Update 2)

[BiG StroOnZ]

In case you missed it, NVIDIA 2 weeks ago had posted a video clip from its GeForce account on YouTube with the caption, "Something super is coming..." Despite the fact that not much was revealed in the 16-second clip:

Quote

The most recent rumor traces back to a post on China's Weibo forum. The usual disclaimer applies (take this with a grain of salt and a pound of skepticism), but according to the post, NVIDIA is planning to release "Super" versions of its GeForce RTX 2080, 2070, and 2060 graphics cards.

 

Here are the rumored specs of each one:

 

• GeForce RTX 2080 Super: 3,072 CUDA cores, TU104-450 GPU, 8GB GDDR6 (16Gbps)
• GeForce RTX 2070 Super: 2,560 CUDA cores, TU104-410 GPU, 8GB GDDR6 (14Gbps)
• GeForce RTX 2060 Super: 2,176 CUDA cores, TU106-410 GPU, 8GB GDDR6 (14Gbps)

 

If the rumor is true, the new Super variants would sport more CUDA cores than the regular versions that are out now, and depending on the model, additional memory that is clocked faster to boot. Here's how the the Super models stack up against the current models:

 

• GeForce RTX 2080 Ti: 4,352 CUDA cores, TU102-300 GPU, 11GB GDDR6 (14Gbps)
• GeForce RTX 2080 Super: 3,072 CUDA cores, TU104-450 GPU, 8GB GDDR6 (16Gbps)
• GeForce RTX 2080: 2,944 CUDA cores, TU104-400 GPU, 8GB GDDR6 (14Gbps)
• GeForce RTX 2070 Super: 2,560 CUDA cores, TU104-410 GPU, 8GB GDDR6 (14Gbps)
• GeForce RTX 2070: 2,304 CUDA cores, TU106-410 GPU, 8GB GDDR6 (14Gbps)
• GeForce RTX 2060 Super: 2,176 CUDA cores, TU106-410 GPU, 8GB GDDR6 (14Gbps)
• GeForce RTX 2060: 1,920 CUDA cores, TU106-200 GPU, 6GB GDDR6 (14Gbps)

 

There's no mention of a GeForce RTX 2080 Ti Super model, but we've included the regular variant in the above for posterity. As it pertains to the GeForce RTX 2080, the Super version would have more CUDA cores and faster memory (16Gbps instead of 14Gbps), whereas the GeForce RTX 2070 Super would just have more CUDA cores. At the bottom of the stack, the GeForce RTX 2060 Super would get a bump from 6GB to 8GB of GDDR6 memory.

 

What's also interesting is that the GeForce RTX 2060 Super is said to be using the full TU106-410 GPU (as is the 2080 Super using the full TU104). Not only will it have more CUDA cores, but the full TU106-410 GPU supports a wider 256-bit memory bus, up from 192-bit on the regular model.

 

Source: https://hothardware.com/news/nvidia-super-geforce-rtx-turing-lineup-specs-leaked

 

I suspected in the original thread, and suggested, that these "SUPER" cards most likely were going to be cards with more CUDA Cores, and faster GDDR6. We'll have to see how it plays out, but this is the most sensible route for a refresh of Turing. We saw some Pascal cards that had faster memory, as well as a card like the 1070 Ti (with more CCs) which slotted in nicely between the 1070 and 1080; making the concept not too outlandish. Aside from that, it is purported that these "SUPER" Turing refreshes will launch sometime in July. Therefore, I imagine as we get closer to that date, we should see some more leaks; conceivably even performance leaks or more concrete specifications.  

 

Little update:

Quote

GeForce RTX 20 SUPER: Announcement next week, to launch mid-July.

Spoiler

 

 

 

Our sources have also referred to next week as a possible announcement of the new series. What’s important, however, the GeForce RTX 20 Super series will not launch till mid-July. NVIDIA has only just started talking to AIBs about the new series, and the time for a Turing refresh preparation is obviously still required. We do not expect significant changes to existing custom designs, as all SUPER SKUs will use existing GPUs (just different variants)

 

We have also confirmed the GPU variants that will launch with SUPER series. The GeForce RTX 2080 SUPER will feature Turing TU104-450 GPU. This alone points towards 8GB GDDR6 memory, which for this very SKU will be faster (16 Gbps vs 14 Gbps) than the rest of the RTX stack. At this point, there is no doubt that this model will have all CUDA cores enabled (3072 to be exact). Meanwhile, the RTX 2070, which was used by AMD during Radeon RX 5700 XT presentation, will receive a faster variant with an additional 256 CUDA cores. This card will be powered by TU104-410 GPU. The RTX 2060 Super will basically be a cut-down version of RTX 2070 non-Super, featuring TU106-410 GPU with full 256-bit memory bus. This means 8GB GDDR6 memory, which is a 2GB more than RTX 2060 non-Super.

To give credit where it’s due, Igor Wallossek from Tom’s Hardware Germany was the first to report the correct variants:

Spoiler

 

 

 

 

 

 

 

Source 2: https://videocardz.com/81046/nvidia-geforce-rtx-20-super-series-to-launch-mid-july

 

 

As we get closer to the supposed reveal of these products, we have an update on the "expected" prices of the SUPER cards:

 

Quote

• RTX 2080 SUPER - $799
• RTX 2070 SUPER - $599
• RTX 2060 SUPER - $349-429

Here's what an update to the prices of the "old" pre-SUPER variants, combined with the new SUPER cards, would look like (according to TweakTown):

• GeForce RTX 2080 SUPER - $799
• GeForce RTX 2080 - $549-$600 (possible price drop price)
• GeForce RTX 2070 SUPER - $599
• GeForce RTX 2070 - $449-499 (possible price drop price)
• GeForce RTX 2060 SUPER - $349-429
• GeForce RTX 2060 - $299 (possible price drop price)

 

Source 3: https://www.tweaktown.com/news/66332/nvidia-geforce-rtx-super-pricing-799-599-429/index.html

 

 

[Fasauceome]

Just now, BiG StroOnZ said:

take this with a grain of salt and a pound of skepticism

how about a pound of salt and a truckload of skepticism? considering the Ryzen rumors anyway.

[Fnige]

Oooooh look,

 

COULDVE CALLED IT "TI" ^{Except for 2080 but we dont need 3 types of 2080s right}

[19_blackie_73]

And the purpose of these cards would be... what?

[Firewrath9]

WTF how would a TU-104-410 have both 2560 and 2944 cuda cores at the same time???

 

[Princess Luna]

nVidia once again shows their gigantic middle finger to early adopters [:

[BiG StroOnZ]

27 minutes ago, Firewrath9 said:

WTF how would a TU-104-410 have both 2560 and 2944 cuda cores at the same time???

 

2944 CCs is the original 2080, the 2560 CCs is for the 2070 Super (the original 2070 had only 2304 CCs). You can scale back CCs as much as you want (usually in specific increments). And if you never released a "full fat" or fully enabled chip, then of course you have the ability to release products with more CCs (but obviously not beyond what the chip is capable of; in the case of TU104, it maxes out at 3,072 CCs). 

[Firewrath9]

42 minutes ago, BiG StroOnZ said:

2944 CCs is the original 2080, the 2560 CCs is for the 2070 Super (the original 2070 had only 2304 CCs). You can scale back CCs as much as you want (usually in specific increments). And if you never released a "full fat" or fully enabled chip, then of course you have the ability to release products with more CCs (but obviously not beyond what the chip is capable of; in the case of TU104, it maxes out at 3,072 CCs). 

if its "scaled back" then it gets a different designation, i.e. 1060 3GB vs 1060 6gb, 1070 vs 1070 ti, 1080 ti vs Titan Xp vs Titan X (Pascal), 1050 vs 1050 ti, Quadro P400 vs P600, P620 vs P1000, etc etc

[BiG StroOnZ]

15 minutes ago, Firewrath9 said:

if its "scaled back" then it gets a different designation, i.e. 1060 3GB vs 1060 6gb, 1070 vs 1070 ti, 1080 ti vs Titan Xp vs Titan X (Pascal), 1050 vs 1050 ti, Quadro P400 vs P600, P620 vs P1000, etc etc

 

Oh I think I see what you're saying/concern is. The 2070 Super is using a scaled back TU104 GPU (the original 2080 also uses a scaled back TU104, however the 2080 Super uses the fully enabled TU104), whereas the original 2070 is using the fully enabled TU106 GPU. 

[Firewrath9]

9 minutes ago, BiG StroOnZ said:

 

Oh I think I see what you're saying/concern is. The 2070 Super is using a scaled back TU104 GPU (the original 2080 also uses a scaled back TU104, however the 2080 Super uses the fully enabled TU104), whereas the original 2070 is using the fully enabled TU106 GPU. 

What I'm saying is why are the chip designation for the 2080 vs 2070 Super the same?

TU104-401 for both, whilst other chips that differ in CUDA count (lets say 1050 vs 1050 ti) use different designations (GP107-300-A1 vs GP107-400-A1

[BiG StroOnZ]

On 6/7/2019 at 6:35 PM, Firewrath9 said:

What I'm saying is why are the chip designation for the 2080 vs 2070 Super the same? TU104-401 for both, whilst other chips that differ in CUDA count (lets say 1050 vs 1050 ti) use different designations (GP107-300-A1 vs GP107-400-A1

 

Nothing too complicated. NVIDIA probably has many not so desirable 2080 TU104 dies available (mediocre or average silicon yields), that while can't make it as RTX 2080 parts, can make it as cut down parts for their "2070 Super."  Explains why the 2080 Super is a -450, while the 2080 and 2070 Super are -410. 

[Firewrath9]

4 minutes ago, BiG StroOnZ said:

 

Nothing too complicated. NVIDIA probably has many not so desirable 2080 TU104 GPU dies available (mediocre or average silicon yields), that while can't make it as RTX 2080 parts, can make it as cut down parts for their "2070 Super." Explains why the 2080 Super is a -450, while the 2080 and 2070 Super are -410. 

no 2080 and 2070 cannot use the same designation even if they are the exact same piece of silicon i.e. all previous nvidia chips

[BiG StroOnZ]

48 minutes ago, Firewrath9 said:

no 2080 and 2070 cannot use the same designation even if they are the exact same piece of silicon i.e. all previous nvidia chips

 

Well, here's the odd thing that makes this scenario different. Again though, this is speculation. Nonetheless, the RTX 2070 is using completely different silicon than the RTX 2070 Super (if the rumors are correct). The 2070 Super is a completely different GPU. The original 2070 uses a TU106, while the 2070 Super supposedly uses TU104. Therefore, already, we are in unorthodox territory. There's no way to release a TU106 part with more CUDA Cores, since the original 2070 is already maxed out. Meaning, only alternative way to make a faster "xx70" series card, would mean to use a cut down TU104. The most sensible and cost-effective way to do this, would be to use GPU dies that couldn't make the cut as regular RTX 2080's. In this case, whatever GPU the RTX 2080 is currently using; that being the TU104-410.

[illegalwater]

Sounds like a pretty lame refresh, hopefully it comes with a price cut too.

[rawrdaysgoby]

Nvda's branding is getting on the level of bad, of intel's level 

[Brooksie359]

This sounds like some pointless GPU launch if it is real. 

[thorhammerz]

47 minutes ago, Firewrath9 said:

no 2080 and 2070 cannot use the same designation even if they are the exact same piece of silicon i.e. all previous nvidia chips

Well, the only possible explanations are either:

• a typo
• inaccuracy / speculation of the leak itself
• NVIDIA themselves deciding to fuck away the uniqueness of their code names.

We'll know for sure in a few weeks / months (or whenever these things actually release) .

[descendency]

Cant wait for the RTX Titan SUPER TI

[Guest]

Sorry Nvidia but this is just complete crap.

[Firewrath9]

45 minutes ago, BiG StroOnZ said:

 

Well, here's the odd thing that makes this scenario different. Again though, this is speculation. Nonetheless, the RTX 2070 is using completely different silicon than the RTX 2070 Super (if the rumors are correct). The 2070 Super is a completely different GPU. The original 2070 uses a TU106, while the 2070 Super supposedly uses TU104. Therefore, already, we are in unorthodox territory. There's no way to release a TU106 part with more CUDA Cores, since the original 2070 is already maxed out. Meaning, only alternative way to make a faster "xx70" series card, would mean to use a cut down TU104. The most sensible and cost-effective way to do this, would be to use GPU dies that couldn't make the cut as regular RTX 2080's. In this case, whatever GPU the RTX 2080 is currently using; that being the TU104-410.

if they would use the TU104 die, they would give ti a new designation, i.e. TU104-310 or something like that.

therefore:

16 minutes ago, thorhammerz said:

Well, the only possible explanations are either:

• a typo
• inaccuracy / speculation of the leak itself
• NVIDIA themselves deciding to fuck away the uniqueness of their code names.

We'll know for sure in a few weeks / months (or whenever these things actually release) .

 

[BiG StroOnZ]

On 6/7/2019 at 7:47 PM, Firewrath9 said:

if they would use the TU104 die, they would give ti a new designation, i.e. TU104-310 or something like that.

therefore:

 

I agree with @thorhammerz too, however, I believe the point the writer of the article was trying to simply get across: was that the 2070 Super would be using TU104 silicon instead of TU106 silicon. While it happens to be that the specific nomenclature is not yet currently detailed, still, all that is known currently; is that it is the same silicon as the 2080. It would be fitting that it gets a new suffix moniker - unless of course, as stated, NVIDIA is doing away with unique identifications for GPU variants. What I do find interesting though, is that the other cards (the 2080 Super and the 2060 Super) have new and specific designations already outlined. Which makes the topic of it simply being a typo or inaccuracy (TU104-410) a bit perplexing. On the other hand, it isn't the first time an author has copied information verbatim from a leaker or source (regardless of potential accuracy or inaccuracy).  

 

Update: It does look like it was a simple typo as the original RTX 2080 uses a TU104-400 GPU not a -410. 

[PocketNerd]

LOL Bait-And-Switch!

"Here's some cards that are super awesome!"
*Not even a year later*

PSYCH! These weren't even their final form!

[Results45]

Can't Nvidia get the naming sensible, short, and sweet for once?

 

How about.........

• RTX 2080TI-S
• RTX 2080S
• RTX 2070TI-S
• RTX 2070S
• RTX 2060S

Or just "RTS" for "Ray-tracing Supercharged" (or Ray-Traced Shooter) 

.

[corsairian]

well looks like they left the stack open for another 2080 TI.

 

GeForce RTX 2080 Super TI: 3712 Cuda Cores... ( half the cuda cores of the 2080ti vs the 2080 Super)

 

Lol I could see this happening in 4 months just in time for Christmas.

[thorhammerz]

59 minutes ago, Results45 said:

Can't Nvidia get the naming sensible, short, and sweet for once?

 

How about.........

• RTX 2080TI-S

.

Now introducing the RTX 2080TI-KS, with an all-clock boost clock of 2000 MHz right out of the box!